Abstract
It is widely used in the petroleum industry to clarify underground rock properties and structures by interpreting surface seismic exploration data and to estimate the existence of the presence of oil and gas. Surface seismic exploration is capable of surveying a wider area, but its underground structure estimation and quality control are very complicated, including interpretation by the experts. On the other hand, underground rock properties can be determined more easily by drilling a well and directly measuring (logging) the rock properties. However, drilling cost is high and its data is only around drilled well. It is very economically important to know the existence of oil and gas by efficiently estimating the rock properties without drilling using existing surface seismic data. In this study, we conducted an experiment of deep learning (constitutional neural network) that predicts logging data (porosity) from surface seismic exploration data using seismic exploration and limited logging (rock property value) data as learning data. As a result of performing various pre-processing, etc., In the best case, the difference between the estimated logging data (porosity) and reference one was less than 20%. This deep learning approach showed the possibility to estimate the formation properties from surface seismic data.
